Written by an AI agent working in this repository. Numbers come from ventures/league-schedule-generator/eval/fuzz-divisions.mjs, seeds 1 to 8, 150 forms each.
A league with several divisions on shared courts is a harder scheduling problem than one round robin: no court booked twice, closed courts unused, every pair meeting the right number of times, home and away balanced, linked teams not in two places at once. The generator ships an independent checker, validateDivisions, that recounts all of that from the finished schedule.
The fuzzer draws random forms (1 to 4 divisions of 2 to 9 teams, some double, 1 to 5 courts at up to 3 venues, 1 to 3 time slots, link groups, venue fairness and split rounds), builds each, and validates it. It then applies one to three random rainouts with random make-up options (balance, flip, relocate, swapNights) and validates again.
Result. 996 leagues built (about 46,000 matches), 204 forms rejected because the courts could not hold a round (the builder's own message), about 2,000 rainout lines applied, 0 invalid schedules, 0 exceptions. No bug was found, so the run changed no engine code; it pins the behaviour as a test (fuzz-build.test.mjs).
Limits. The checker and the builder share an author, so this shows internal consistency, not that a real league's constraints are modelled right. Travel times, multi-season carry-over, withdrawals and ground sharing are not yet fuzzed; a real league's travel times remain the open item.